On Mon, Jan 26, 2015 at 2:13 PM, Brian Rak <brak@xxxxxxxxxxxxxxx> wrote: > I have an existing cluster where all the hosts were just added directly, for > example: > > # ceph osd tree > # id weight type name up/down reweight > -1 60.06 root default > ... > -14 1.82 host OSD75 > 12 1.82 osd.12 up 1 > -15 1.82 host OSD80 > 13 1.82 osd.13 up 1 > -16 1.82 host OSD83 > 14 1.82 osd.14 up 1 > -17 1.82 host OSD78 > 15 1.82 osd.15 up 1 > -18 1.82 host OSD82 > 17 1.82 osd.17 up 1 > -19 1.82 host OSD84 > 16 1.82 osd.16 up 1 > > > Ultimately, I'm trying to reconfigure this so I can withstand a failure of a > rack, without losing data. I found http://dachary.org/?p=2536 which seemed > to be pretty helpful in setting this up. > > I have the OSDs set to pull their location via 'osd crush location hook' > script. This appears to work properly, as the output properly lists the > configured rack: > > === osd.20 === > create-or-move updated item name 'osd.20' weight 1.82 at location > {host=OSD81,rack=rack1,root=default} to crush map > Starting Ceph osd.20 on OSD81... > > However, I don't see OSD81 appear under the rack in the 'ceph osd tree' > output. The only way I can get this to happen is if I manually move the > OSD, with 'ceph osd crush move OSD81 rack=rack1'. > > Am I wrong in expecting that 'create-or-move' would handle this without > manual intervention? If so, what moves will 'create-or-move' handle > automatically? I have enough nodes that manually moving them all to the > proper place would be time consuming. create-or-move creates the osd within the most specific bucket specified (host, in your case), and will move the osd if it exists in a different bucket, but it doesn't move any of the buckets to match. You'll have to invoke crush move for those. -Greg _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |